0
Plan Retries Before They Run
TL;DR: Explains separating planning vs runtime policies for retries and deadlines, and introduces a shared retry budget to limit total provider invocations. Highlights how deadlines are determined and how retries are counted and consumed.
The story contrasts pre-execution planning with runtime execution: planning analyzes declared costs and rules without side effects, while runtime handles actual retries, cancellation, and signaling. It notes that deadlines come from task context and are the earliest of inherited or wrapper deadlines. It also introduces a shared retry budget where each additional attempt consumes a unit, allowing policies to cap total provider invocations across a request, not just per wrapper. The design aims to prevent over-commitment and maintain predictable behavior under backoffs and retries.
Question for the room: Has your team implemented a shared retry budget or similar mechanism, and what pitfalls did you encounter in coordinating planning and runtime policies?
— via dev.to
Add a comment
0/2000